Applied Scientist II · Amazon
Masum Billah
Applied Scientist building agentic AI systems, on-device LLM runtimes, and intelligent sensing platforms.
About
Researcher, engineer, and builder of intelligent systems.
I'm an Applied Scientist II at Amazon, where I architect and deploy multi-agent LLM systems that power Alexa+ — Amazon's next-generation agentic AI assistant. My work spans autonomous reasoning pipelines, on-device AI runtimes, tool-use orchestration frameworks, and evaluation systems operating at terabyte scale.
Before Amazon, I completed my PhD in Computer Science at the University of Virginia, where I pioneered research at the intersection of deep learning, wireless sensing, and IoT systems. My work has been published in top-tier venues including ACM SenSys, IPSN, and BuildSys, and has earned multiple Best Paper Awards.
I'm driven by the challenge of building AI systems that reason autonomously, adapt in real-time, and operate reliably at production scale — from cloud infrastructure to edge devices. My research and engineering span the full stack: from LLM planning architectures and multi-agent coordination to on-device inference optimization and ambient intelligence platforms.
Education
PhD, Computer Science
University of Virginia
2018 — 2023
MS, Computer Science
University of Virginia
2018 — 2022
BS, Computer Science
Bangladesh University of Engineering & Technology
2013 — 2017
Experience
Building production AI systems at scale.
Amazon
Applied Scientist II
- •Designed and deployed multi-agent LLM systems for autonomous reasoning and evaluation at TB-scale
- •Built on-device agentic runtime architectures for Alexa+ with planning and execution models
- •Developed orchestration frameworks with validation, retries, and multi-turn reasoning pipelines
- •Delivered 10+ production ML systems powering Alexa ambient intelligence experiences
- •Promoted to Applied Scientist II for sustained high-impact technical contributions
Apple
Machine Learning Research Intern
- •Built ML system for automated handwashing detection deployed to 2M+ Apple Watch users
- •Worked on real-world sensing systems combining on-device ML with large-scale data processing
Reve Systems
Software Engineer, NLP Team
- •Developed applications including a digital screen reader and real-time sign language detection
Projects
From agentic AI systems to production-scale infrastructure.
Alexa+ Agentic Runtime Architecture
Two-model agentic architecture separating planning and execution for on-device AI assistants. Features orchestration with dependency resolution, retries, validation, and multi-turn reasoning. Optimized for edge deployment using shared SmolLM2 models with lightweight LoRA adapters.
Agentic AutoML Research System
Multi-agent LLM system that autonomously analyzes datasets, generates algorithms, evaluates performance, and iteratively improves solutions. Handles TB-scale data and is deployed internally for 50+ researchers at Amazon.
Syntara AI
End-to-end LLM-powered financial intelligence platform with real-time pipelines monitoring 13.5k+ stocks and crypto assets. Features AI reasoning pipelines synthesizing multi-source data into actionable insights with interactive follow-up.
Distributed LLM Infrastructure
Led a team of 10 engineers to build a distributed LLM system across heterogeneous hardware. Designed scalable orchestration using Docker, Redis, and monitoring systems for efficient large-scale training and inference.
RF-based Occupancy Detection
BLE-based system using Deep Q-Network (DQN) for indoor occupancy detection. Deployed on resource-constrained embedded devices with robust performance using wireless signal features.
SolarWalk
ML system to identify occupants using solar cell voltage traces. An unobtrusive sensing approach without cameras or wearables for privacy-preserving occupant identification.
Multimodal IoT Localization & AR
Combined images and wireless signals for accurate device localization in Augmented Reality. Enables intuitive point-and-control interaction with IoT devices.
Video Caption Generation
Encoder-decoder architecture for generating captions from unseen video clips. Leverages multimodal representation learning for vision-language understanding.
Publications
15+ peer-reviewed papers in top-tier venues.
Sensei: Empowering LLMs as Self-Learning Sensing Experts
M. Billah, A. Agrawal, M. Bocca
fReeLoaders: An IoT Ecosystem for Real-Time Deadline-Driven Task Scheduling using RL
M. Clyburn, M. Billah et al.
Technical Skills
Full-stack AI expertise from research to production.
Agentic AI & LLM Systems
AI Systems & Infrastructure
Machine Learning
Backend & Data Systems
Languages
Awards & Patents
Recognition for research impact and invention.
Awards & Honors
Amazon Inventor Award
Amazon · 2024
Amazon Patent Incentive Award
Amazon · 2024
Best Paper Award — ACM/IEEE SEC 2025
ACM/IEEE · 2025
Best Paper Award — ACM BuildSys Workshop (DFHS)
ACM · 2019
University of Virginia Endowed Fellowship
UVA · 2022–2023
University of Virginia CS Fellowship
UVA · 2018–2019
Patents
Associating Devices using Wireless Signals
Amazon · 2024
Device Automatic Grouping with Audio and Wireless Signal Multimodal Fusion
Amazon · 2025
Get in Touch
Open to collaborations, research discussions, and interesting problems.